Struct RuntimeClient 

pub struct RuntimeClient { /* private fields */ }

HTTP access to the serving runtime.

Tokenization always goes through the runtime unless a local tokenizer is configured, so the bridge never needs a copy of the model’s vocabulary. Rendering goes through it too unless a local renderer is configured.

Implementations

impl RuntimeClient

pub fn new( client: Client, base: impl Into<String>, model: impl Into<String>, api_key: Option<String>, ) -> Self

Point a client at the runtime’s native base URL.

pub async fn apply_template( &self, messages: &[ChatMessage], chat_template_kwargs: &Value, ) -> Result<String>

Render messages with the runtime’s own chat template.

pub async fn tokenize(&self, content: &str) -> Result<Vec<u32>>

Tokenize without special tokens, matching fastjev’s encode_prompt.

pub async fn detokenize(&self, tokens: &[u32]) -> Result<String>

Decode tokens for the answer-slot round-trip check.